Inline quality assurance for Sheet Molding Compound (SMC) manufacturing faces a bottleneck: there are not enough large-scale, accurately labeled image datasets to train machine-learning models for defect detection. Without such data, vision-based inspection systems are difficult to develop and validate in production settings.
The paper introduces a process-informed, photorealistic digital twin designed to address this scarcity. By simulating the C-SMC strip deposition process, the twin can generate realistic images that reflect the actual manufacturing conditions. These synthetic images could serve as training data for machine-learning models, reducing the need for manually annotated real-world images.
If effective, this approach would make inline quality assurance more practical for SMC production. The abstract does not provide details on validation or performance, so the claims remain at the proposal stage. Still, the idea of using a process-aware simulator to fill a data gap is a notable direction for manufacturing inspection.